Impact of ENSO 2016–17 on regional climate and malaria vector dynamics in Tanzania

Large scale modes of climate variability, including the El Niño Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD), have been shown to significantly impact mosquito-borne diseases in the Tropics, including malaria. However, the mechanistic cascade from ENSO and the IOD, to induced changes in regional climate and ultimately mosquito abundance and behaviour is poorly understood. Mosquito population dynamics, behaviour and their potential to transmit disease are all sensitive to micro-climatic conditions. The warm phase of ENSO (El Niño) tends to be associated with increased precipitation and outbreaks of various vector-borne diseases, while the cold phase (La Niña) can cause drought during the short rains over East Africa. The sensitivity of Anopheles mosquito population dynamics and host-seeking behaviour to ENSO and to the resulting micro-climatic conditions, were investigated in the Kilombero Valley in Tanzania. From June 2016 to September 2017, changes in the timing and intensity of the rainy seasons and temperature due to the ENSO 2016–17 were observed. Mosquitoes were collected using Centres for Disease Control and Prevention (CDC) light traps indoors and mosquito electrocuting traps in- and outdoors. Changes in abundance and biting behaviour of Anopheles arabiensis and Anopheles funestus were correlated with climate and micro-climate. The impacts of El Niño on climate and mosquito abundance were not clear. However, the study area experienced a drought due to La Niña during which both vector species declined significantly. An. arabiensis densities stayed more stable at higher temperatures and were found in higher numbers outdoors with respect to An. funestus. For both species, indoor temperature and season determined their host-seeking location, with higher temperatures and the wet season driving them outside. The study confirmed the influence of ENSO and micro-climate on malaria vector abundance and host-seeking behaviour, generating hypotheses for predicting the impact of future ENSO on malaria risk and vector control. Our observation of higher outdoor biting during warmer conditions indicates that indoor vector control strategies may become proportionally less effective during this time.


Introduction
Despite successful control efforts and a vast reduction in cases and deaths over the last decade, malaria is still a major public health concern in many parts of the world (Bhatt et al 2015a, 2015b, WHO 2018. Over 90% of all malaria cases and deaths occur in sub-Saharan Africa, and malaria continues to be the most significant mosquito-borne disease hampering public health and socio-economic development in this region. The World Health Organisation estimated that there was an increase of two million cases between 2016 and 2017 globally (WHO 2018) and climate is considered a possible contributor. In most endemic regions of sub-Saharan Africa, mosquitoes of the Anopheles gambiae (An. gambiae s.s. and An. arabiensis its sibling species) and Anopheles funestus complex are the primary vectors for malaria (Collins and Besansky 1994, Donnelly et al 2001, Sinka et al 2012. There has been widespread and consistent demonstration of strong association between seasonal precipitation and abundance of these vectors because of the dependence of their aquatic larval stages on standing water (Lindblade et al 1999, Oesterholt et al 2006, Zhou et al 2007, Kelly-Hope et al 2009, Bomblies 2012. Additionally, other micro-climatic variables such as temperature have significant impacts on several aspects of adult vector fitness, behaviour and transmission potential Lindsay 2003, Kulkarni et al 2006). For example, the gonotrophic cycle of mosquitoes (e.g. time between biting and laying eggs) shortens and adult survivorship increases with temperature up to a thermal tolerance threshold, beyond which their fitness is impaired (Paaijmans et al 2010). Additionally, the sporogonic development rate of malaria parasites within vectors increases with temperature, thus increasing their transmission potential (Shapiro et al 2017). These mosquito demographic and epidemiological parameters ultimately determine rates of human exposure to infected mosquito bites.
Micro-climatic and seasonal environmental variation can also impact human exposure to malaria in another way: by altering the timing and location where vectors bite. Currently, malaria vector control in Africa is primarily conducted through application of insecticides inside houses (Hemingway 2014). This is based on use of Long lasting Insecticidal Nets (LLINs) and Indoor Residual Spraying (IRS); both of which are very successful in reducing malaria in Africa (Bhatt et al 2015a(Bhatt et al , 2015b. The success of LLINs is based on their ability to exploit the behavioural predisposition of African malaria vectors to primarily feed on humans (anthropophagy) during sleeping hours, inside houses (endophagy), and rest indoors after feeding (endophily) Ferguson 2009, Killeen et al 2017). These behaviours increase the probability of vectors coming into contact with insecticides either during host-seeking (e.g. LLINs) or resting on walls after blood feeding (e.g. IRS). Consequently, the upscaling of these control measures has coincided with a substantial decrease in malaria vector abundance (Bayoh et  The host-seeking and resting behaviour of malaria vectors has previously been shown to be influenced by the micro-climate of their immediate environment Thomas 2011, Ngowo et al 2017). However, the relationship between larger-scale climate phenomena such as the El Niño Southern Oscillation (ENSO), micro-climate and mosquito host-seeking behaviour are less clear. Understanding the mechanistic cascade from the ENSO, to induced changes in regional climate, and ultimately mosquito abundance and host-seeking behaviour is important in times of a changing climate and increasing insecticide resistance. Indoor-based control methods may be increasingly challenged both by insecticide resistance and climate-driven changes in vector behaviour.
The Kilombero Valley in southern Tanzania has experienced historically high malaria transmission with 226 infective bites per person per year in 2012 (Lwetoijera et al 2014). Since then, it has seen a reduction in infective bites to 15.9 ib/p/yr in 2015 (Kaindoa et al 2017b, Finda et al 2018 due to control measures (LLINs and IRS). The availability of high quality historical data on vector ecology and transmission, and a high coverage with LLINs make this valley a good model for quantifying impacts of extreme climate events in areas of Africa were transmission has also been declining.
Worldwide the warm phase of ENSO, El Niño, is associated with the movement of warm waters from western part to the eastern part of the Pacific Ocean. El Niño has been associated with infectious disease outbreaks, including Rift Valley fever, malaria, and cholera; increased risk of arbovirus and malaria transmission in Latin America and Southeast Asia; and outbreaks of malaria and cholera in India (Hales et al 1999, Chretien et al 2015, Anyamba et al 2019. El Niño leads to a warming of the atmosphere in the Tropics which can last several months to a year after the event (Tyrrell et al 2015). El Niño often causes flooding over eastern Africa during the short rainy season (Oct to Dec), while its influence on the long rains (March-May) is less clear (Nicholson 2017). The effect of ENSO on malaria incidence in East Africa significantly varies regionally. During the 1958-59 El Niño, very conducive climate conditions resulted in three million additional malaria cases in the highlands of Ethiopia (Fontaine et al 1961). During the 1997-98 El Niño, higher temperatures and increased precipitation resulted in increased malaria prevalence in the highlands and north-eastern Kenya (Brown et al 1998). In contrast, an overall reduction of malaria cases was reported in the Usambara mountains of Tanzania during the 1997-98 El Niño which was attributed to heavy rainfall washing away mosquito breeding sites (Lindsay et al 2000). Conversely, an increase in malaria cases was reported at lower elevations for two other locations (Kagera and Morogoro) in Tanzania that year (Carlstedt et al 2004). The cold phase of ENSO, La Niña, tend to be associated with colder and drier conditions over East Africa (Omumbo et al 2011).
The relationship between La Niña, regional climate anomalies and malaria burden has not been extensively studied.
The Indian Ocean dipole (IOD) is an oscillation of sea surface temperatures (SSTs) in which the western part of the Indian Ocean becomes alternately warmer and then colder than the eastern part of the Indian Ocean. The positive phase of the IOD (when the western part is warmer than the eastern part of the Indian Ocean) has also been associated with flood conditions during the short rains (October-December) over Eastern Africa (Behera et al 2005). A positive phase of the IOD tends to increase easterlies crossing the Indian Ocean, bringing more moisture to eastern Africa during the short rains.
To fully understand the impacts of these climate anomalies on vector-borne diseases, thorough surveillance of vectors and their behaviour through all phases is needed. This study therefore aims to primarily determine the effects of ENSO 2016-17 on malaria vector abundance and host-seeking behaviour, as a means to understand the potential impact of these events on malaria transmission and to inform control strategies.

Experimental design
Entomological surveillance was carried out to investigate associations between climate variables and vector abundance, species composition and biting behaviour (biting time and location) between June 2016 and September 2017. Malaria vectors were repeatedly sampled in each village at the same four households for four consecutive days each month. On the first day of sampling, an index house was selected in each village on the basis of being accessible, and the presence and willingness of residents to participate. Three additional houses were recruited in the vicinity of the index house to achieve the required sample size (4 households), with houses being within 100-200 m of one another. In each set of houses, two houses were selected where livestock were kept (e.g. goats or cattle), and two without, because of the known impact on vector species composition and location of biting.

Trapping methodology
Two trapping methods were used to sample hostseeking mosquitoes throughout the study. CDC Miniature light traps were used to collect mosquitoes host seeking indoors at night. Additionally we introduced a relatively new sampling method; mosquito electrocuting traps (MET) which provide an exposure-free method to directly measure mosquito landing rates on people in indoor and outdoor settings. METs can be used to sample mosquitoes attempting to feed on a human volunteer from 6 pm to 6 am in indoor and outdoor settings (Maliti et al 2015, Govella et al 2016. In June 2016 host-seeking mosquitoes were collected with CDC light traps in four houses for four nights in each village. From July 2016, CDC light traps were used in three out of the four houses, with METs being used at the remaining house (one indoors and one outdoors). Trap types were rotated each night following a Latin square design. These methods were selected because Centers of Disease Control and Prevention (CDC) light traps provide a widely used proxy of overall mosquito abundance and indoor biting rates (Briët et al 2015), while MET traps give information on hourly biting time and location. CDC light traps were deployed from 6 pm to 6 am every night by placing them approximately 1.5 m above ground and close to the foot of a bed in which between one and four people were sleeping under a LLIN. Collections with METs were also conducted from 6 pm to 6 am. The MET is composed of four electrified panels positioned in a square surrounding the lower legs of a seated volunteer, that intercept and kill mosquitoes on approach, while the rest of the volunteer's body is protected by netting. Each hour, the MET was turned off for 15 min to allow mosquitoes caught on the surface to be removed, recorded and stored. At the house allocated for MET collection, one trap was positioned within a living room and another outside (∼5 m from house), on each night as described elsewhere (Govella et al 2016). The volunteers sitting in the MET traps were swapped between indoor and outdoor trapping stations every hour to minimise bias due to differing attractiveness to mosquitoes. Additional data on mosquito abundance and species composition based on CDC light trap collections (indoors) from a previous study (2012 in the same villages (350 households) were used as a baseline for comparison with non-El Niño years.

Mosquito identification and molecular analyses
All mosquitoes collected in traps were killed by chloroform. The number and sex of those morphologically identified as belonging to the An. gambiae s.l or An. funestus s.l. complex or Culex species were recorded (Edwards 1941, Gillies and De Meillon 1968, Gillies and Coetzee 1987. A subset of An. gambiae s.l. collected (n=5600, 22% of total) were identified to species level by polymerase chain reaction (PCR) (Scott et al 1993). For this, 5 individual mosquitoes were sampled from each trap per night. Mosquitoes were sampled for indoor and outdoor MET separately. With an amplification rate of 92.5%, laboratory results confirmed them all to be An. arabiensis. On this basis of the predominance of An. arabiensis in the An.
gambiae s.l. tested here and in other concurrent studies in the area (Govella et al 2009, Marsden et al 2014, Maliti et al 2015, Kaindoa et al 2017a, all An. gambiae s.l. collected were assumed to be, An. arabiensis. PCR analysis was also conducted on members of the Anopheles funestus s.l. to identify them to species level (n=2104, 20% of total, amplification rate 87.6%) (Koekemoer et al 2002). The majority of An. funestus s. l. specimens were identified to be An. funestus funestus (97%) followed by An. rivulorum (1.4%) and An. funestus lessonii (1.1%). Additionally, mosquitoes were pooled in batches of a maximum of 10 per sampling tube per trap type, per night (An. arabiensis: n=14 700, 59% of total and An. funestus: n=7890, 75% of total) Enzyme Linked Immunosorbent Assays were used to test for presence of Plasmodium malaria parasites (Beier et al 1990).

Environmental data
Indoor temperature and humidity were recorded with Tiny Tag Plus 2 data loggers (Gemini data loggers, UK, Ltd) placed inside houses on each night of sampling (approximately 1 m above the ground). These data were used to calculate the indoor saturation deficit for each house using established methods (Allen et al 1998). Season was defined for each sampling month as wet or dry depending on the monthly amount of rainfall. A month with rainfall over 1 mm per day on average was defined as 'wet'.
Daily climate data was retrieved from different sources. Daily rainfall (mm) and temperature (°C) were obtained from the Ifakara GloBe weather station (GRWS 100 Campbell Scientific) installed at the Ifakara Health Institute (IHI) (8.114 17°S, 36.674 84°E ) within the floodplain (see figure 1). The weather station has been recording from 18th November 2014. To calculate anomalies (e.g. departure from the long term means), we utilised gridded climate data. Daily rainfall data from the Climate Hazards group Infrared Precipitation with Stations (CHIRPS) dataset at 0.05°× 0.05°spatial resolution was used for the period 1981-2017 (Funk et al 2015). Monthly gridded temperature data (0.5°× 0.5°resolution) which combines weather station data from the Global Historical Climatology Network version 2 with the Climate Anomaly Monitoring System was utilised for the same period (Fan and van den Dool 2008). Monthly anomalies were calculated with respect to the 1981-2017 period for the gridded products. A comparison between gridded and weather station data is provided for rainfall on figure S1 is available online at stacks.iop.org/ ERL/14/075009/mmedia and for temperature on figure S2. Time variability is well reproduced by the gridded data; but both CAMS and CHIRPS data tend to overestimate temperature and rainfall over Ifakara. The Nino 3.4 index data (calculated as monthly sea surface temperature anomalies with respect to 1981-2010 climatology over the region 5°N-5°S and 170-120°W) and the Dipole Mode Index (calculated as the difference between the average SST in the region 50-70°E and 10°S-10°N minus the average SST in the box 90-110°E and 10°S-0°N) based on the HadISST data (Rayner et al 2003) were downloaded from KNMI climate explorer (Trouet and van Oldenborgh 2013).

Ethics
Before the study began, meetings were held with community leaders in all villages during which they were informed about the purpose of the study and their participation requested. After their permission had been granted, the study team visited each village and informed consent was obtained from each head of household where mosquito trapping was conducted. The study was previously approved by the Ifakara Health Institutional Review Board (Institutional Ethics Clearance: Certificate number IHI/IRB/No: 037-2016). It was further approved by the University of Liverpool ethics board (RETH001036).

Analysis and models
The potential environmental drivers of vector abundance and host-seeking location (indoors versus outdoors) were investigated in generalised linear mixed models (GLMMs) that included explanatory variables of nightly minimum, mean and maximum temperature (in°C) and relative humidity (RH in%) indoors, saturation deficit indoors (in kPa) and season (wet or dry). Effects of temperature, humidity and saturation deficit on mosquito abundance and host-seeking location were investigated using generalised linear mixed models with the 'glmmTMB' package in R statistical software (Brooks et al 2017). Mosquito abundance was estimated as the mean number of vectors caught per CDC light trap per night and in MET per hour. In all models, all micro-climatic variables and season were fitted as fixed effects while household id, date and trap number were fitted as random effects. Model selection was conducted using the Akaike Information Criterion (AIC), by sequentially selecting models with lower AIC values and the rule of parsimony (Bolker et al 2009). For the models on host-seeking location, the hourly number of mosquitoes collected by MET was fitted as the response variable, while trap location (inor outdoors) was fitted as a two-way interaction term to all fixed effects. Hour of collection was included as random effect nested in house. Data was modelled as following a negative binomial distribution due to the degree of overdispersion in the data (using a test for overdispersion by (Cameron and Trivedi 1990).

Mosquito bionomics
A total of 28 799 mosquitoes were collected using CDC light traps during 778 trap nights across the study (table 1). A further 9061 mosquitoes were collected with METs (combined indoors and outdoors), across 215 trap nights. With both methods, more than twice as many An. arabiensis were caught than An. funestus, with the majority of collected female mosquitoes unfed (table 1). Infection rates with Plasmodium falciparum were 0.013% for An. arabiensis and 0.025% in the An. funestus.

ENSO, regional climate anomalies and mosquito dynamics
The 2015-16 El Niño was one of the strongest events on record. This event started in October-November 2014, peaked during the boreal winter 2015 before declining during the boreal spring 2016. This warm event was followed by a mild La Niña signal from June 2016 to January 2017 (figures 2(a) and S3(a)). The SST signal in the Indian Ocean was not very clear in 2016; however, a moderate positive phase of the IOD occurred in 2017 ( figure 2(a)). In Ifakara, rainfall tends to occur from November until May, with a peak in March-April (figures 2(b) and S3(c)). During the study period from June 2016 to September 2017, mean monthly temperature oscillated between 23°C and 30°C (figures 2(b) and S3(b)). The warmest months are usually between October and January ( figure S3(b)). On average, the positive (negative) phase of ENSO, El Niño (La Niña), is associated with increased (decreased) rainfall conditions over eastern Africa during the short rains ( figure S4(g)). The positive (negative) phase of the IOD is also associated with increased (decreased) precipitation over Tanzania ( figure S4(h)). The relationship between rainfall in Tanzania and the IOD is even more pronounced than ENSO during the short rains as shown by Behera et al (2005). One of the wettest short rain season occurred in 1997 in Tanzania, when both positive phases of the IOD and ENSO co-occurred ( figure S4(b)). In 2015, wetter than average conditions were observed over Tanzania ( figure S4(c)), however, 2015 was not as wet as 1997. La Niña events in 1982, 1997 and 2016 were mostly related to drier than average conditions in Tanzania (figures S4(d)-(f)). The relationship between ENSO, DMI and land temperature in Tanzania is not significant and not clear, when long term trends are removed ( figure S5). However, colder than average conditions are associated with the positive phase of the DMI (figure S5(h)) over northern Tanzania. This temperature signal is consistent with increased rainfall conditions ( figure S4(h)) which tend to cool the land surface.
Anopheles arabiensis density peaked in phase during the February-April rainy reason, with An. funestus peaking 2-3 months later (May-August with a peak in July, see figures 2(c) and S3(d)). Despite some differences across study sites, this feature was relatively robust around the Kilombero Valley ( figure S6). Climate anomalies (e.g. departures from the long term mean), warmer (colder) temperatures were experienced during El Niño (La Niña) events in Ifakara ( figure S7(b)). During El Niño 2015-16, more rainfall was observed over the region (figure S7(c)). Conversely, a significant drought occurred during the following La Niña between October 2016 and February 2017 ( figure S7(c)). These results for Ifakara are consistent with the aforementioned findings at country scale (figures S4 and S5).
Populations of both Anopheles vector species crashed below detection during the drought associated with La Niña (October 2016-February 2017). Drought conditions (figure S7(c)) and lower abundance (figure S7(d)) were previously observed from January to March 2012. The relative population crash was more pronounced for An. funestus than An. arabiensis ( figure S7(d)). Lagged monthly correlations between temperature and mosquito abundance were not significant using a standard Pearson test except when temperature was leading An. arabiensis abundance by 1, 4 and 5 months (figures S8(a) and S8(b)). However, rainfall was significantly positively correlated with An. funestus at a 2 months lag (r=0.64, p<0.001, figure S4(d)).
3.3. Effects of micro-climate on host seeking location and abundance More An. arabiensis were caught outdoors than indoors (figures 3(a), (c) and S9) while the opposite was observed for An. funestus (figures 3(b), (d) and S9(b)). No mosquitoes were caught when night-time relative humidity dropped below 40% (figure 3), with vector abundance highest when RH>60%. Very few An. funestus were caught when mean temperature exceeded 32°C (figures 3(b) and (d)). In contrast, the abundance of An. arabiensis was relatively stable between 24°C and 32°C, and the maximum suitable temperature was found to be 32°C ( figure 3(a)).
The mean number of An. arabiensis collected per CDC trap per night was negatively associated with maximum indoor temperature and showed a positive association with the wet season (table 2(a)). The abundance of An. funestus catches was negatively associated with increasing temperature (minimum and maximum, table 2(b)).
Exophily, defined as the relative proportion of mosquitoes caught in outdoor versus indoors METs, was predicted to increase in the wet season for both vector species outdoors of the total MET catch (table 2(c)).
An. arabiensis and An. funestus exhibited differences in the timing of their nightly host seeking ( figure 4). Specifically, An. arabiensis, was very active from early evening (18.00) until midnight, mostly outdoors ( figure 4(a)). Biting rates of An. arabiensis decreased after midnight following the observed decrease in temperature (figure 4(c)) and increase in RH (figure 4(d)). A secondary peak in An. arabiensis biting activity was observed in the early morning hours, when temperature increases again (06.00). Conversely, An. funestus, which was mostly caught indoors, was most active in the middle of the night between 23.00 and 01.00 ( figure 4(b)).

Discussion
In the face of climate change impacts on East Africa, understanding the effects of climate (synoptic scale) and micro-climate on disease vector abundance and behaviour is essential. We investigated the relationship between ENSO, regional climate and micro-climate on two major malaria vector populations, An. arabiensis and An. funestus, in Tanzania.
Because of the juxtaposition of ENSO onto global warming in 2016-17, the Kilombero Valley in Tanzania, an area of historically high malaria transmission, has seen changes in the timing and intensity of the rainy seasons and temperature. In turn, the effect of ENSO on the malaria vector populations in the Kilombero Valley proved to be complex and vector specific.
The relationship between the 2015-16 El Niño, regional climate anomalies and the Anopheles population in the study area was not straightforward. However, we highlighted a robust relationship between the following La Niña, regional drought conditions and the crash of the vector population in the area. Such sudden and strong decline in mosquito numbers has not been observed in the region during years with normal dry season conditions (Ngowo et al 2017). The sudden decline was more pronounced for An. funestus which also consistently showed a more limited tolerance to high temperature conditions than An. arabiensis. However, data was limited and baseline data was derived from information collected throughout the previous 3+years only. Due to this relatively short, but high quality record, extended by our study, continuing to monitor mosquito dynamics would be beneficial.
Almost twice as many An. arabiensis as An. funestus were caught with all trap types. In line with other studies, malaria vector populations showed strong seasonality (Koenraadt et al 2004, Ngowo et al 2017. Notably, An. arabiensis numbers peaked in phase with rainfall, while An. funestus numbers were highest 2-3 months after the rainfall peak as seen in the previous year 2015 (Ngowo et al 2017). This is most likely related to their respective larval ecology-An. funestus develops much slower than An. arabiensis (Kirby andLindsay 2009, Lyons et al 2013) and prefers still, clean, more permanent water bodies, while An. arabiensis happily breeds in temporary water bodies (Gillies and De Meillon 1968, Minakawa et al 1999, Charlwood et al 2000, Gimnig et al 2001. During the rainy season, breeding sites become turbid, while water bodies become an oasis of still and clean water during the dry season-making them highly attractive for oviposition and increasing the survival of An. funestus' offspring, increasing their density. Laboratory studies also indicate a negative effect of temperatures above 28°C and fluctuating temperatures on An. funestus' larval development and adult survival (Charlwood 2017). The peak rainy season in Tanzania is associated with both, high temperature and temperature fluctuations.
In the Kilombero Valley region, overall temperature seems favourable year round for Anopheles and the relationship could not be interpreted between monthly changes in temperature and mosquito abundance, but it had a marked effect on the micro-climate level. At a local scale, higher temperatures inside houses were associated with decreased abundance of anophelines (table 2(a)). In our study, the highest temperature at which An. funestus were caught indoors was 32°C which is in line with published maximum threshold of survival for anophelines  proved to be more resilient to high temperatures and was even collected at 36°C. Unsurprisingly, An. arabiensis numbers were positively associated with the wet season, while there was a negative relationship with maximum temperature. The apparently higher sensitivity to micro-climatic conditions by An. funestus was confirmed by our GLMM model showing negative associations with both minimum and maximum temperature. Other variables such as village, RH and saturation deficit did not seem to influence host-seeking mosquito abundance indoors, even though both populations decreased dramatically during the period of drought. The lack of significant effect of RH could point to the existence of thresholds which were not captured by our model type. The influence of micro-climate on the location of host-seeking vectors as measured by the METs, showed an increase of the proportion of anophelines caught outdoors during the wet season. An. arabiensis and An. funestus exhibited significant differences in their night-time host-seeking behaviour. An. arabiensis was caught at much higher numbers outdoors and in the early evening hours.
While indoor numbers gradually decreased throughout the night, outdoor numbers showed another peak at midnight. The majority of An. funestus on the other hand, was consistently caught indoors.
The known, more flexible behaviour of An. arabiensis (Fornadel et al 2010, Russell et al 2011, Gordicho et al 2014, including increased outdoor feeding earlier in the evening Norris 2013, Kaindoa et al 2017a), has serious implications for the success of vector control. Mosquitoes that bite outdoors and earlier in the evening, before people get under their LLIN, as reported by other studies (Tirados et al 2006, Maliti 2016, tend to avoid the main control strategy used against them. On the other hand, Anopheles funestus was found to be highly endophilic in our study and this is confirmed by others (Gillies and De Meillon 1968, Pates and Curtis 2005, Lounibos 2007) which makes it more vulnerable to LLINs and most likely caused its decline in many areas after the introduction of control methods (Meyrowitsch et al 2011, Zhou et al 2011. With ENSO now occurring in a warmer background, due to

Conclusion
While the effect of El Niño could not be established in this study, La Niña caused drought at a regional scale which led to decreased mosquito abundance. In terms of micro-climate, our study confirmed that temperature and to a lesser extent RH has an impact on vector behaviour, with new control strategies for outdoor biting vectors early in the evening urgently needed. try of Science and Education and Ministry of Economic Cooperation and Development.
We also thank Dr Andy Hardy from Aberystwyth University, UK for the preparation of the study site map.
The authors declare no conflict of interest.

Data
The weather station data can be requested by contacting Mrs Fatuma Matwewe at the Ifakara Health Institute, Tanzania (fmatwewe@ihi.or.tz). The gridded climate data is publicly available online. The mosquito abundance data is publicly available on the NERC's Environmental Information Data Centre at https:// catalogue.ceh.ac.uk/documents/89406b06-d0aa-4120-84db-a5f91b616053.